octave-maintainers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 4.2 Release Checklist


From: Marco Caliari
Subject: Re: 4.2 Release Checklist
Date: Wed, 31 Aug 2016 14:07:08 +0200 (CEST)
User-agent: Alpine 2.10 (DEB 1266 2009-07-14)

Dear Rik, dear maintainers,

Cristiano Dorigo worked with SOCIS and the official end of programme is
tomorrow. He worked a lot with pcg, gmres, bicg, bicgstab, cgs fixing
several errors, bugs, incompatibilities and not always clear documentations.
For better compatibility, he changed some output diagnostic (value of the
flag error, number of iterations) and also the result itself (for instance,
when the methods reach the maximum number of iterations, Matlab returns the
best approximation, aka with minimal residual, not the latest as Octave
does).

I followed him weekly, I think he made a great job, but honestly it is quite
hard to me to make a deep review (maybe alone) of all his work in the next
six days. And a deep review is needed, since he touched almost all the
iterative methods for linear systems. Is it possible to have more time for
the review? Anyone else would like to help me in the review? Is it possible
that we introduce in 4.2 only the new pcg and gmres (probably the most used
solvers, and the first two Cristiano started to study and modify)?


Hi Marco

If the changes he made are that deep and widespread it's probably not a
good idea to merge them so close to the release.  We probably should have
been reviewing and merging his work as it was done so it was not all done
at the end.

Yes, this is my fault.

Your proposal of merging only his earlier work may be better if it's already
reviewed and tested.

We are doing it now. If Friday is the deadline, I will inform you what was revised up to Friday. I'm a little bit worried about the fact that Cristiano modified, in some cases, the error flags for compatibility with Matlab. The side effect is that some working codes, expecting some types of failures, could not work anymore. Would it be possible to use the linear-algebra package for keeping the current legacy codes? Or the other way round, for putting the new codes?

Marco



reply via email to

[Prev in Thread] Current Thread [Next in Thread]